summaryrefslogtreecommitdiffstats
path: root/libFDK/src/arm/dct_arm.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'libFDK/src/arm/dct_arm.cpp')
-rw-r--r--libFDK/src/arm/dct_arm.cpp395
1 files changed, 395 insertions, 0 deletions
diff --git a/libFDK/src/arm/dct_arm.cpp b/libFDK/src/arm/dct_arm.cpp
new file mode 100644
index 0000000..dd0ca09
--- /dev/null
+++ b/libFDK/src/arm/dct_arm.cpp
@@ -0,0 +1,395 @@
+/****************************************************************************
+
+ (C) copyright Fraunhofer IIS (2004)
+ All Rights Reserved
+
+ Please be advised that this software and/or program delivery is
+ Confidential Information of Fraunhofer and subject to and covered by the
+
+ Fraunhofer IIS Software Evaluation Agreement
+ between Google Inc. and Fraunhofer
+ effective and in full force since March 1, 2012.
+
+ You may use this software and/or program only under the terms and
+ conditions described in the above mentioned Fraunhofer IIS Software
+ Evaluation Agreement. Any other and/or further use requires a separate agreement.
+
+
+
+ $Id$
+
+***************************************************************************/
+
+
+
+#ifdef FUNCTION_dct_IV_func1
+
+/*
+ Note: This assembler routine is here, because the ARM926 compiler does
+ not encode the inline assembler with optimal speed.
+ With this version, we save 2 cycles per loop iteration.
+*/
+
+__asm void dct_IV_func1(
+ int i,
+ const FIXP_SPK *twiddle,
+ FIXP_DBL *RESTRICT pDat_0,
+ FIXP_DBL *RESTRICT pDat_1)
+{
+ /* Register map:
+ r0 i
+ r1 twiddle
+ r2 pDat_0
+ r3 pDat_1
+ r4 accu1
+ r5 accu2
+ r6 accu3
+ r7 accu4
+ r8 val_tw
+ r9 accuX
+ */
+ PUSH {r4-r9}
+
+ /* 44 cycles for 2 iterations = 22 cycles/iteration */
+dct_IV_loop1_start
+/* First iteration */
+ LDR r8, [r1], #4 // val_tw = *twiddle++;
+ LDR r5, [r2, #0] // accu2 = pDat_0[0]
+ LDR r4, [r3, #0] // accu1 = pDat_1[0]
+
+ SMULWT r9, r5, r8 // accuX = accu2*val_tw.l
+ SMULWB r5, r5, r8 // accu2 = accu2*val_tw.h
+ RSB r9, r9, #0 // accuX =-accu2*val_tw.l
+ SMLAWT r5, r4, r8, r5 // accu2 = accu2*val_tw.h + accu1*val_tw.l
+ SMLAWB r4, r4, r8, r9 // accu1 = accu1*val_tw.h - accu2*val_tw.l
+
+ LDR r8, [r1], #4 // val_tw = *twiddle++;
+ LDR r7, [r3, #-4] // accu4 = pDat_1[-1]
+ LDR r6, [r2, #4] // accu3 = pDat_0[1]
+
+ SMULWB r9, r7, r8 // accuX = accu4*val_tw.h
+ SMULWT r7, r7, r8 // accu4 = accu4*val_tw.l
+ RSB r9, r9, #0 // accuX =-accu4*val_tw.h
+ SMLAWB r7, r6, r8, r7 // accu4 = accu4*val_tw.l+accu3*val_tw.h
+ SMLAWT r6, r6, r8, r9 // accu3 = accu3*val_tw.l-accu4*val_tw.h
+
+ STR r5, [r2], #4 // *pDat_0++ = accu2
+ STR r4, [r2], #4 // *pDat_0++ = accu1
+ STR r6, [r3], #-4 // *pDat_1-- = accu3
+ STR r7, [r3], #-4 // *pDat_1-- = accu4
+
+/* Second iteration */
+ LDR r8, [r1], #4 // val_tw = *twiddle++;
+ LDR r5, [r2, #0] // accu2 = pDat_0[0]
+ LDR r4, [r3, #0] // accu1 = pDat_1[0]
+
+ SMULWT r9, r5, r8 // accuX = accu2*val_tw.l
+ SMULWB r5, r5, r8 // accu2 = accu2*val_tw.h
+ RSB r9, r9, #0 // accuX =-accu2*val_tw.l
+ SMLAWT r5, r4, r8, r5 // accu2 = accu2*val_tw.h + accu1*val_tw.l
+ SMLAWB r4, r4, r8, r9 // accu1 = accu1*val_tw.h - accu2*val_tw.l
+
+ LDR r8, [r1], #4 // val_tw = *twiddle++;
+ LDR r7, [r3, #-4] // accu4 = pDat_1[-1]
+ LDR r6, [r2, #4] // accu3 = pDat_0[1]
+
+ SMULWB r9, r7, r8 // accuX = accu4*val_tw.h
+ SMULWT r7, r7, r8 // accu4 = accu4*val_tw.l
+ RSB r9, r9, #0 // accuX =-accu4*val_tw.h
+ SMLAWB r7, r6, r8, r7 // accu4 = accu4*val_tw.l+accu3*val_tw.h
+ SMLAWT r6, r6, r8, r9 // accu3 = accu3*val_tw.l-accu4*val_tw.h
+
+ STR r5, [r2], #4 // *pDat_0++ = accu2
+ STR r4, [r2], #4 // *pDat_0++ = accu1
+ STR r6, [r3], #-4 // *pDat_1-- = accu3
+ STR r7, [r3], #-4 // *pDat_1-- = accu4
+
+ SUBS r0, r0, #1
+ BNE dct_IV_loop1_start
+
+ POP {r4-r9}
+
+ BX lr
+}
+
+#endif /* FUNCTION_dct_IV_func1 */
+
+
+#ifdef FUNCTION_dct_IV_func2
+
+FDK_INLINE
+/* __attribute__((noinline)) */
+static void dct_IV_func2(
+ int i,
+ const FIXP_SPK *twiddle,
+ FIXP_DBL *pDat_0,
+ FIXP_DBL *pDat_1,
+ int inc)
+{
+ FIXP_DBL accu1, accu2, accu3, accu4, accuX;
+ LONG val_tw;
+
+ accu1 = pDat_1[-2];
+ accu2 = pDat_1[-1];
+
+ *--pDat_1 = -(pDat_0[1]>>1);
+ *pDat_0++ = (pDat_0[0]>>1);
+
+ twiddle += inc;
+
+__asm
+ {
+ LDR val_tw, [twiddle], inc, LSL #2 // val_tw = *twiddle; twiddle += inc
+ B dct_IV_loop2_2nd_part
+
+ /* 42 cycles for 2 iterations = 21 cycles/iteration */
+dct_IV_loop2:
+ SMULWT accuX, accu2, val_tw
+ SMULWB accu2, accu2, val_tw
+ RSB accuX, accuX, #0
+ SMLAWB accuX, accu1, val_tw, accuX
+ SMLAWT accu2, accu1, val_tw, accu2
+ STR accuX, [pDat_0], #4
+ STR accu2, [pDat_1, #-4] !
+
+ LDR accu4, [pDat_0, #4]
+ LDR accu3, [pDat_0]
+ SMULWB accuX, accu4, val_tw
+ SMULWT accu4, accu4, val_tw
+ RSB accuX, accuX, #0
+ SMLAWT accuX, accu3, val_tw, accuX
+ SMLAWB accu4, accu3, val_tw, accu4
+
+ LDR accu1, [pDat_1, #-8]
+ LDR accu2, [pDat_1, #-4]
+
+ LDR val_tw, [twiddle], inc, LSL #2 // val_tw = *twiddle; twiddle += inc
+
+ STR accuX, [pDat_1, #-4] !
+ STR accu4, [pDat_0], #4
+
+dct_IV_loop2_2nd_part:
+ SMULWT accuX, accu2, val_tw
+ SMULWB accu2, accu2, val_tw
+ RSB accuX, accuX, #0
+ SMLAWB accuX, accu1, val_tw, accuX
+ SMLAWT accu2, accu1, val_tw, accu2
+ STR accuX, [pDat_0], #4
+ STR accu2, [pDat_1, #-4] !
+
+ LDR accu4, [pDat_0, #4]
+ LDR accu3, [pDat_0]
+ SMULWB accuX, accu4, val_tw
+ SMULWT accu4, accu4, val_tw
+ RSB accuX, accuX, #0
+ SMLAWT accuX, accu3, val_tw, accuX
+ SMLAWB accu4, accu3, val_tw, accu4
+
+ LDR accu1, [pDat_1, #-8]
+ LDR accu2, [pDat_1, #-4]
+
+ STR accuX, [pDat_1, #-4] !
+ STR accu4, [pDat_0], #4
+
+ LDR val_tw, [twiddle], inc, LSL #2 // val_tw = *twiddle; twiddle += inc
+
+ SUBS i, i, #1
+ BNE dct_IV_loop2
+ }
+
+ /* Last Sin and Cos value pair are the same */
+ accu1 = fMultDiv2(accu1, WTC(0x5a82799a));
+ accu2 = fMultDiv2(accu2, WTC(0x5a82799a));
+
+ *--pDat_1 = accu1 + accu2;
+ *pDat_0++ = accu1 - accu2;
+}
+#endif /* FUNCTION_dct_IV_func2 */
+
+
+#ifdef FUNCTION_dst_IV_func1
+
+__asm void dst_IV_func1(
+ int i,
+ const FIXP_SPK *twiddle,
+ FIXP_DBL *pDat_0,
+ FIXP_DBL *pDat_1)
+{
+ /* Register map:
+ r0 i
+ r1 twiddle
+ r2 pDat_0
+ r3 pDat_1
+ r4 accu1
+ r5 accu2
+ r6 accu3
+ r7 accu4
+ r8 val_tw
+ r9 accuX
+ */
+ PUSH {r4-r9}
+
+dst_IV_loop1
+ LDR r8, [r1], #4 // val_tw = *twiddle++
+ LDR r5, [r2] // accu2 = pDat_0[0]
+ LDR r6, [r2, #4] // accu3 = pDat_0[1]
+ RSB r5, r5, #0 // accu2 = -accu2
+ SMULWT r9, r5, r8 // accuX = (-accu2)*val_tw.l
+ LDR r4, [r3, #-4] // accu1 = pDat_1[-1]
+ RSB r9, r9, #0 // accuX = -(-accu2)*val_tw.l
+ SMLAWB r9, r4, r8, r9 // accuX = accu1*val_tw.h-(-accu2)*val_tw.l
+ SMULWT r4, r4, r8 // accu1 = accu1*val_tw.l
+ LDR r7, [r3, #-8] // accu4 = pDat_1[-2]
+ SMLAWB r5, r5, r8, r4 // accu2 = (-accu2)*val_tw.t+accu1*val_tw.l
+ LDR r8, [r1], #4 // val_tw = *twiddle++
+ STR r5, [r2], #4 // *pDat_0++ = accu2
+ STR r9, [r2], #4 // *pDat_0++ = accu1 (accuX)
+ RSB r7, r7, #0 // accu4 = -accu4
+ SMULWB r5, r7, r8 // accu2 = (-accu4)*val_tw.h
+ SMULWB r4, r6, r8 // accu1 = (-accu4)*val_tw.l
+ RSB r5, r5, #0 // accu2 = -(-accu4)*val_tw.h
+ SMLAWT r6, r6, r8, r5 // accu3 = (-accu4)*val_tw.l-(-accu3)*val_tw.h
+ SMLAWT r7, r7, r8, r4 // accu4 = (-accu3)*val_tw.l+(-accu4)*val_tw.h
+ STR r6, [r3, #-4] ! // *--pDat_1 = accu3
+ STR r7, [r3, #-4] ! // *--pDat_1 = accu4
+
+ LDR r8, [r1], #4 // val_tw = *twiddle++
+ LDR r5, [r2] // accu2 = pDat_0[0]
+ LDR r6, [r2, #4] // accu3 = pDat_0[1]
+ RSB r5, r5, #0 // accu2 = -accu2
+ SMULWT r9, r5, r8 // accuX = (-accu2)*val_tw.l
+ LDR r4, [r3, #-4] // accu1 = pDat_1[-1]
+ RSB r9, r9, #0 // accuX = -(-accu2)*val_tw.l
+ SMLAWB r9, r4, r8, r9 // accuX = accu1*val_tw.h-(-accu2)*val_tw.l
+ SMULWT r4, r4, r8 // accu1 = accu1*val_tw.l
+ LDR r7, [r3, #-8] // accu4 = pDat_1[-2]
+ SMLAWB r5, r5, r8, r4 // accu2 = (-accu2)*val_tw.t+accu1*val_tw.l
+ LDR r8, [r1], #4 // val_tw = *twiddle++
+ STR r5, [r2], #4 // *pDat_0++ = accu2
+ STR r9, [r2], #4 // *pDat_0++ = accu1 (accuX)
+ RSB r7, r7, #0 // accu4 = -accu4
+ SMULWB r5, r7, r8 // accu2 = (-accu4)*val_tw.h
+ SMULWB r4, r6, r8 // accu1 = (-accu4)*val_tw.l
+ RSB r5, r5, #0 // accu2 = -(-accu4)*val_tw.h
+ SMLAWT r6, r6, r8, r5 // accu3 = (-accu4)*val_tw.l-(-accu3)*val_tw.h
+ SMLAWT r7, r7, r8, r4 // accu4 = (-accu3)*val_tw.l+(-accu4)*val_tw.h
+ STR r6, [r3, #-4] ! // *--pDat_1 = accu3
+ STR r7, [r3, #-4] ! // *--pDat_1 = accu4
+
+ SUBS r0, r0, #4 // i-= 4
+ BNE dst_IV_loop1
+
+ POP {r4-r9}
+ BX lr
+}
+#endif /* FUNCTION_dst_IV_func1 */
+
+#ifdef FUNCTION_dst_IV_func2
+
+FDK_INLINE
+/* __attribute__((noinline)) */
+static void dst_IV_func2(
+ int i,
+ const FIXP_SPK *twiddle,
+ FIXP_DBL *RESTRICT pDat_0,
+ FIXP_DBL *RESTRICT pDat_1,
+ int inc)
+{
+ FIXP_DBL accu1,accu2,accu3,accu4;
+ LONG val_tw;
+
+ accu4 = pDat_0[0];
+ accu3 = pDat_0[1];
+ accu4 >>= 1;
+ accu3 >>= 1;
+ accu4 = -accu4;
+
+ accu1 = pDat_1[-1];
+ accu2 = pDat_1[0];
+
+ *pDat_0++ = accu3;
+ *pDat_1-- = accu4;
+
+
+ __asm
+ {
+ B dst_IV_loop2_2nd_part
+
+ /* 50 cycles for 2 iterations = 25 cycles/iteration */
+
+dst_IV_loop2:
+
+ LDR val_tw, [twiddle], inc, LSL #2 // val_tw = *twiddle; twiddle += inc
+
+ RSB accu2, accu2, #0 // accu2 = -accu2
+ RSB accu1, accu1, #0 // accu1 = -accu1
+ SMULWT accu3, accu2, val_tw // accu3 = (-accu2)*val_tw.l
+ SMULWT accu4, accu1, val_tw // accu4 = (-accu1)*val_tw.l
+ RSB accu3, accu3, #0 // accu3 = -accu2*val_tw.l
+ SMLAWB accu1, accu1, val_tw, accu3 // accu1 = -accu1*val_tw.h-(-accu2)*val_tw.l
+ SMLAWB accu2, accu2, val_tw, accu4 // accu2 = (-accu1)*val_tw.l+(-accu2)*val_tw.h
+ STR accu1, [pDat_1], #-4 // *pDat_1-- = accu1
+ STR accu2, [pDat_0], #4 // *pDat_0++ = accu2
+
+ LDR accu4, [pDat_0] // accu4 = pDat_0[0]
+ LDR accu3, [pDat_0, #4] // accu3 = pDat_0[1]
+
+ RSB accu4, accu4, #0 // accu4 = -accu4
+ RSB accu3, accu3, #0 // accu3 = -accu3
+
+ SMULWB accu1, accu3, val_tw // accu1 = (-accu3)*val_tw.h
+ SMULWT accu2, accu3, val_tw // accu2 = (-accu3)*val_tw.l
+ RSB accu1, accu1, #0 // accu1 = -(-accu3)*val_tw.h
+ SMLAWT accu3, accu4, val_tw, accu1 // accu3 = (-accu4)*val_tw.l-(-accu3)*val_tw.h
+ SMLAWB accu4, accu4, val_tw, accu2 // accu4 = (-accu3)*val_tw.l+(-accu4)*val_tw.h
+
+ LDR accu1, [pDat_1, #-4] // accu1 = pDat_1[-1]
+ LDR accu2, [pDat_1] // accu2 = pDat_1[0]
+
+ STR accu3, [pDat_0], #4 // *pDat_0++ = accu3
+ STR accu4, [pDat_1], #-4 // *pDat_1-- = accu4
+
+dst_IV_loop2_2nd_part:
+
+ LDR val_tw, [twiddle], inc, LSL #2 // val_tw = *twiddle; twiddle += inc
+
+ RSB accu2, accu2, #0 // accu2 = -accu2
+ RSB accu1, accu1, #0 // accu1 = -accu1
+ SMULWT accu3, accu2, val_tw // accu3 = (-accu2)*val_tw.l
+ SMULWT accu4, accu1, val_tw // accu4 = (-accu1)*val_tw.l
+ RSB accu3, accu3, #0 // accu3 = -accu2*val_tw.l
+ SMLAWB accu1, accu1, val_tw, accu3 // accu1 = -accu1*val_tw.h-(-accu2)*val_tw.l
+ SMLAWB accu2, accu2, val_tw, accu4 // accu2 = (-accu1)*val_tw.l+(-accu2)*val_tw.h
+ STR accu1, [pDat_1], #-4 // *pDat_1-- = accu1
+ STR accu2, [pDat_0], #4 // *pDat_0++ = accu2
+
+ LDR accu4, [pDat_0] // accu4 = pDat_0[0]
+ LDR accu3, [pDat_0, #4] // accu3 = pDat_0[1]
+
+ RSB accu4, accu4, #0 // accu4 = -accu4
+ RSB accu3, accu3, #0 // accu3 = -accu3
+
+ SMULWB accu1, accu3, val_tw // accu1 = (-accu3)*val_tw.h
+ SMULWT accu2, accu3, val_tw // accu2 = (-accu3)*val_tw.l
+ RSB accu1, accu1, #0 // accu1 = -(-accu3)*val_tw.h
+ SMLAWT accu3, accu4, val_tw, accu1 // accu3 = (-accu4)*val_tw.l-(-accu3)*val_tw.h
+ SMLAWB accu4, accu4, val_tw, accu2 // accu4 = (-accu3)*val_tw.l+(-accu4)*val_tw.h
+
+ LDR accu1, [pDat_1, #-4] // accu1 = pDat_1[-1]
+ LDR accu2, [pDat_1] // accu2 = pDat_1[0]
+
+ STR accu3, [pDat_0], #4 // *pDat_0++ = accu3
+ STR accu4, [pDat_1], #-4 // *pDat_1-- = accu4
+
+ SUBS i, i, #1
+ BNE dst_IV_loop2
+ }
+
+ /* Last Sin and Cos value pair are the same */
+ accu1 = fMultDiv2(-accu1, WTC(0x5a82799a));
+ accu2 = fMultDiv2(-accu2, WTC(0x5a82799a));
+
+ *pDat_0 = accu1 + accu2;
+ *pDat_1 = accu1 - accu2;
+}
+#endif /* FUNCTION_dst_IV_func2 */